In her 35 years of teaching, Julie Brown has never seen anything like it. Fake students — bots, as she calls them — have been enrolling in her courses at an alarming rate. Last fall, the marketing professor at Santa Barbara City College dropped 30 of her 140 students from an online Business 101 course. She also removed eight potential bots from an in-person introductory management course. This spring, she estimates about 10 of her 45 students across five courses may have been fraudulent.

Determining who is real and who is not takes many hours of work. Before the term starts, Brown, who teaches up to 8 courses per semester, emails all of her students, informing them they will have assignments due the first week of class. They must upload a photo of themselves, write an essay, and comment on the essays of several other students. She tells them that if they don’t do all of these things, she will drop them.

Once classes begin, she has to follow up. Did every student submit the required essay? How about the photo? Did they respond to their classmates? Did anything seem like it was done with AI? Did her in-person students show up or email an excuse? She writes again: Please respond. Please fix this. Please know that if you don’t do what I’m asking, I will remove you from this class. “I’m obsessed with this,” she says.

Brown knows fraudsters are betting that she will let things slide. After all, real students sometimes choose to not do the work. And life can get in the way. But if she doesn’t push, two very bad things might happen. The scammers could get what they came for: free financial aid, often in the form of Pell Grants. And real students could be prevented from enrolling in her classes. “It breaks my heart,” she says, “when a student comes to me during office hours and shares that they gave up on trying to get into a class that they needed.”

But diligence has taken a toll on her and her colleagues. Time is just one cost. Doubting whether your students are real is another. Worrying that your class might be canceled if most of your students turn out to be fake is a third. But faculty have to build these brick walls, she says, to stop scammers stealing places from real students.

“I feel like a new war against bots begins every new semester,” Brown says.

Five years after the pandemic hit, California community colleges are still battling one of the most nefarious problems to surface after courses moved online en masse. Criminals continue to find ways to evade various checkpoints in their pursuit of financial aid and other benefits that come with being a college student. Several California Republicans in the U.S. House recently asked the Department of Education and the Department of Justice to launch an investigation.

These fraudsters — sometimes called ghost students, Pell runners, or straw students — are hitting the system’s central application portal at an alarming rate. While spam filters stop many of them, colleges are spending hundreds of hours and thousands of dollars finding further fraud. Instructors, as the last line of defense, are struggling to balance due diligence, good teaching practices, and student support.

The fact that California has yet to find a solution raises an unsettling specter. Could this problem become pervasive in other states? Community colleges in Alabama, Illinois, Louisiana, Maryland, Minnesota, New Jersey, and Oregon are already experiencing an influx of bots. (Brown isn’t the only person who uses that shorthand: Administrators and professors often refer to the fraudsters as “bots” in conversation.)

And yet, community colleges by their very nature are handicapped in fighting bad actors looking to gain entry into their system. They pride themselves on their open access and low cost. Every barrier they erect to fight fraud is a potential roadblock for a real student.

AI misuse further complicates matters. California instructors say it can be hard to distinguish a scammer from a student who relies too much on AI. That — combined with the steady growth of online education, where fraudsters most easily hide — puts community colleges in a vulnerable spot.

“They can take quizzes. They can respond to discussions. They can submit work. It’s not as easy to spot as it was before,” says Kim Rich, a professor of criminal justice at Los Angeles Pierce College who has spent more than three years sounding the alarm about bots in classrooms. “It’s becoming more and more difficult every single day.”

When scammers first attacked community colleges in California, they were both harder and easier to detect than they are today. Harder because professors and administrators were overwhelmed by the task of quickly moving their classes online in the midst of a pandemic. Whether students were showing up or doing the work wasn’t their most pressing concern. The federal government was also encouraging students to stay in college, and it disbursed millions of dollars to help institutions keep enrollment up.

But those first fake students were also clumsy, littering their digital trails with easy tells: an email address that was a jumble of numbers and letters. A last name that appeared twice, like “John Smith Smith.” No phone number listed. Often, dozens of applications came in with the same address, sometimes of a vacant house.

It has gone completely out of control.

California’s community-college system was targeted for a number of reasons. It is the largest in the country, enrolling more than 2 million students. It does not require an application fee, removing a barrier to entry. Because the system has a single application portal, called CCCApply, once fake students are in the system, they can attempt to enroll at any number of the state’s 116 campuses. It has very low tuition, which benefits someone who wants to maximize the amount of money they get back on, say, a Pell Grant, after the college gets paid.

Each district also operates independently, making it difficult to create a coordinated response to bot attacks or even know what is happening elsewhere. One campus might have been hit early on, then developed an effective screening mechanism through trial and error. A neighboring institution, unaware of that experience, is caught off guard when they are flooded with fake applications a year later.

Criminals are “throwing as many darts against the wall as they can, and trying to see where they can poke holes,” says Brian Suponcic, senior vice president of sales and client operations of BM Technologies, which offers colleges in California and elsewhere payment systems and fraud-detection services. “Some schools put up walls and some don’t. And the ones that don’t are way more vulnerable.”

Suponcic says there are three basic types of fraud: first-party, which is a single individual aiming to steal money; third-party, which uses other people’s information, often found on the dark web; and synthetic fraud, which creates fake personas from real and fake identifiers. Criminals hope to tap into federal, and sometimes state, financial aid for lower-income students. Based on his company’s data, he says, the average refund a California community-college student receives in a semester ranges between $2,000 to $2,500.

“It’s a quantity game,” he says. “A $14,000 refund might raise a red flag. But $2,000, that’s pretty typical. So I think they try to stay beneath a certain threshold.”

Some fraudsters cause further damage once inside the system — using their .edu email address to conduct phishing scams across campus, for example, says Suponcic. “They’re saying, ‘Hey, there’s a potential problem with your refund or with your tuition payment, you need to give me your login information immediately so that you’re not locked out.’ And unfortunately students fall for that sometimes.”

They can take quizzes. They can respond to discussions. They can submit work. It’s not as easy to spot as it was before.

News that California was besieged by fake students went national in the summer of 2021. The state began asking colleges to file monthly reports about the fraud they were seeing in the application, enrollment, and financial-aid processes. It also provided an additional $75 million for security-network upgrades and anti-fraud technology.

But the fraudsters kept coming.

Santa Barbara City College got hit hard around fall of 2022, says Laura Woyach, co-chair of the accounting and finance department and a member of the academic senate. Discussions happened in pockets, she recalls, with faculty members exchanging information in committee meetings and within the academic senate. Instructors were told to watch their enrollments for bots.

But, says Woyach, “your average professor was not paying attention. They weren’t realizing that this is a computer doing the work. And there was an old adage of like, ‘You have the right to fail. You’ve enrolled in this class, you want to do the work and fail it, that’s fine. Or you want to do a half-assed job, that’s fine. That’s your right.’”

Scammers targeted introductory online asynchronous courses with no prerequisites. If they got blocked in one section of a course, Woyach says, they would skip to another. Within a few semesters they started learning tricks, like checking the “returning student” box because they would receive less scrutiny and be fast-tracked. Administrators, she recalls, struggled to respond.

“They wanted us to start calling and seeing if the phone numbers that were listed were real phone numbers. And that’s where it was just out of control and where I kind of lost it,” Woyach says of the fall 2024 semester. “I was like, I’m sorry, I’m a faculty member. And to have to sit and call a 50-person roster to figure out if these people are real or not? You’re telling me there’s no IT way to to vet these people? Come on.”

Administrators ramped up detection strategies. To strengthen their defenses, the admissions team has identified “a ton” of different characteristics that suggest an enrollee might be fraudulent, says Paloma Arnold, vice president for student affairs. Demonstrating any one of those gets the applicant put on hold for manual review. She says this process is about 95 percent effective.

From a professor’s perspective, at least, the problem has gotten significantly better, says Woyach: “Last year to this year, it has become much more successful.”

But it is hard to stay vigilant. The number of applications jumps right before school starts, since the fraudsters know there is less time to do individual reviews. Accelerated classes and those that start later in the semester are also vulnerable, says María L. Villagómez, vice president for academic affairs. “We can’t dedicate every single week of the semester to fighting bots.”

In the fall of 2024, according to figures the college supplied, Santa Barbara City College received 6,631 new student applications. Of those, 4,288 were blocked by the state’s CCCApply filter. Of the 2,343 applications that made it through, Santa Barbara identified 1,155 as fraud. Even after that, 60 bots got through and were later dropped after enrolling in a course.

In the end, only about one in six applicants to Santa Barbara proved to be real students. “The bots have become incredibly savvy,” says Arnold. “We’re seeing fraudulent applications getting through with very low spam scores, meaning they know how to answer all of the questions correctly. They know all of the things that they’re filtering for in CCCApply and they’ve navigated around it.”

“We need an intervention at this point that is stronger than what we have been able to do locally or even the chancellor’s office,” says Villagómez. “I think we need something coming from lawmakers. Because it has gone completely out of control.”

The arrival of ChatGPT in late 2022 has only made it more difficult for professors to distinguish bots from real students.

One philosophy instructor who teaches for several community colleges in California and asked not to be named because of concerns about job security said that in one college at which fraudsters were overrunning courses, administrators were “gung ho” to incorporate AI into teaching. And while he bans AI use in his classes, it’s clear many students have been using it. He struggled in recent months to determine which students were real.

In a course where he had already dropped a number of enrollees he suspected of being bots, the instructor met by Zoom with a student he believed had relied on AI to do some writing. The student admitted it, and said he was trying to improve the quality of his discussion posts since he felt they were less well-written than his classmates’. Another student whipped through seven assignments in a few minutes and took just 17 minutes to complete an exam, scoring a 92. The professor also asked to speak with her. She never responded.

This semester, the instructor started requiring that every student meet with him by Zoom and complete several assignments within the first three days of class. That, he says, has helped rid his courses of bots and improved student outcomes.

Michele Weber, an associate professor of journalism and communication at Mt. San Jacinto College, teaches several asynchronous courses, including news writing, mass media, and public relations. She has developed strategies for battling bots.

When her students upload their required photos, she does a reverse-image search to see whether they were pulled off the internet. If someone says they are an anthropology or history major, she gives them a second look, because journalism majors are typically the only ones to take certain courses.

When dozens of emails flooded her inbox this year, requesting authorization to join her class after registration had closed, she noted the formulaic writing and the similarities in the reasons given for not registering through the normal process, like a death in the family or the California wildfires. Weber spent a lot of time sifting through the emails, asking for more detail on why the students wanted to join her courses. In the end, the replies she got were so generic that she declined to admit any of them.

In the end, only about one in six applicants to Santa Barbara proved to be real students.

What really unsettles her, though, are the videos. She encountered the first ones last year. In all of her courses, she requires students to upload an introductory video in which they tell her about themselves. The first oddities arrived with dark backgrounds, as if the person had put on a film noir filter. The student’s mouths began moving before sound came out. Watching them, she says, felt like viewing a foreign TV show dubbed into English.

And yet, because she couldn’t say for sure that these were fakes, Weber allowed these students to stay in her courses. Administrators told her to keep monitoring the students just in case they were fraudulent. She ended up dropping three: one who was unresponsive and whose bio didn’t match their video and two who stopped participating after the course-drop date, which was when financial aid was released.

Weber hates that she’s second-guessing everyone. She is angry when she feels like she got duped. She is frustrated not knowing where the lines are between authentic and fake students, especially since she’s also spending even more hours dealing with AI misuse. “I don’t know if they’re real and using ChatGPT or if they’re a bot in my class scamming money. I mean, where is reality here?”

Curry Mitchell, president of the Academic Senate at MiraCosta College and an English professor, has experienced the same feelings. First-week assignments feel more like they’re about weeding out bad actors than getting to know students. When he sees an assignment that looks like it’s written by AI, he says, he responds differently depending on whether he thinks the student is real or not.

Even in-person courses are confusing. If someone emails and says they can’t make the first class, is that a red flag? Or if someone shows up for the first class but not the next two — did they hire a ringer? “It bothers me,” he says. “It’s disturbing that I’m reacting like that sometimes.”

For contingent faculty members, the process of vetting your roster and keeping enrollments above a required minimum can feel like a juggling act.

Akello Stone teaches several introductory sociology courses at El Camino College. As he is getting the semester started, he says, he is simultaneously teaching real students, weeding out fake ones, adding new students from the waitlist, checking to see if those are real people, then getting rid of the fake ones so legitimate students can enroll. But he knows that real students are willing to wait only so long to get into a class.

“In the summer it was really bad,” Stone recalls. “I think I cleared the roster of at least 40 students and had to keep replacing students. I was actually at a point where I was reaching out to former students who had taken other classes of mine and saying, ‘Hey, I have room in this class.’ Because my livelihood is connected to whether or not I have enrollment in my classes.”

Stone is appreciative of his dean, who he says has manually gone through hundreds of names to clear rosters before the semester started so professors didn’t have to.

In such a large, sprawling system, though, it’s no surprise that faculty are responding inconsistently to the bot threat. Some instructors continue to teach in passive ways, professors say, not dropping students who don’t show up or do any work for weeks into the semester, well beyond the period in which they would have secured at least a portion of financial aid.

College administrators say they understand that faculty members don’t have the time, or the motivation, necessarily, to weed out fraud. Contingent instructors, in particular, risk losing income if they diligently remove fraudulent students and then see their course dropped if it falls below an enrollment minimum. “There is a huge financial incentive working against abiding by our policies and processes or even our recommendations,” says Villagómez. “That is hugely problematic.”

Some community colleges offer training for faculty members on how they can help detect fake students. They note that these strategies are also good teaching practices, because they emphasize immediate and sustained engagement.

”We can’t rely on discussion boards and multiple-choice tests,” says Jory Hadsell, vice chancellor for technology and chief technology officer for Foothill-De Anza Community College District. “There have to be ways that students showcase their skills that are not so easily replicated by a bot. And that’s a lot of work faculty are doing.” But it takes time, he says, to encourage every faculty member to change their teaching strategies.

It may be impossible to know the extent of the financial loss caused by these scams, but some news organizations have attempted to add up the cost. EdSource reported last year that since 2021, California community colleges have distributed more than $14 million in aid — mostly federal — that they ultimately wrote off as being fraudulent.

A spokesperson from the California Community Colleges chancellor’s office declined to provide a dollar figure, saying that would require a public-records request. But they said in an email that financial-aid fraud “is extremely low relative to the billions of dollars of state and federal aid disbursed — about 0.21% in the 2023-24 academic year.”

The state has spent about $150 million improving cybersecurity in the past three years. And in early 2024 the community-college system began using ID.me, an optional step that students can take to confirm their identity (although some colleges have seen fraudsters fool that system as well). There is also now a statewide task force for colleges to share information and best practices to detect and mitigate fraud.

The central office is in the process of overhauling CCCApply, which has been criticized for its limited ability to catch fraud. A new application will be ready in the spring of 2026.

In the meantime, the spokesperson said, the system is adding new detection and prevention measures as the fraudsters adapt. “This fight is ongoing and evolves as bad actors continue to shift their attacks and use more sophisticated techniques,” the statement said. “We use a multitude of technological and AI tools available to us to recognize and detect false applications, bots, script attacks, and other evolving threats while verifying identities, designing an improved application process, preventing impersonations, and building in ongoing assessments.”

System officials estimate that 31.4 percent of applications coming into CCCApply in 2024 were fraudulent. That’s up from previous years, although they note that detection has also gotten stronger, so it’s hard to say whether the attacks are actually increasing.

Professors say they would like to see even more done at the local and state level, including better collaboration and data sharing among colleges. Rich, the criminal-justice professor, says the lack of communication has been a real hindrance in developing strategies to fend off bots.

“Some colleges were very progressive and transparent and tried to head off the issues and they did talk to their faculty,” she says. “But it wasn’t public because I think everybody was so worried that the criminals would go, ‘Oh, they did this, we can work around it’. I hate to say it, [the criminals] already knew that. ... They have to be extremely organized, sophisticated, and advanced to be able to do this and not be shut down at this point.”

Last year, Foothill-De Anza Community College District began using a fraud-detection program called LightleapAI, which is being used at 42 community colleges, most of which are in California. N2N Services, the company that runs LightleapAI, is one of several providers aiming to serve colleges overwhelmed by fraud and lacking the manpower or skills to handle it themselves.

“Institutions, they’re spending hundreds of thousands of dollars just to do silly things — like, they took out every application and they go to Google Maps to see if that’s really a home,” says Kiran Kodithala, CEO of N2N. “It’s an extremely cost-prohibitive, extremely inefficient process, and that’s not helping.”

Administrators at El Camino College, where Stone teaches, are also implementing LightleapAI. While bots have been a growing problem on the campus, things were particularly bad this spring. Applications more than doubled — to 21,000 — with about 13,000 determined to be fraudulent. “It’s been hundreds and hundreds and hundreds of hours with our student-services colleagues just to do that identification process,” says Carlos Lopez, vice president for academic affairs.

LightleapAI uses AI pattern recognition to study more than 400 fields within CCCApply and determine what is likely to indicate fraud. The company found that nearly 36 percent of the more than 735,000 applications coming into 36 colleges where LightleapAI was being used this academic year were fraudulent — and that was after CCCApply had done its screening.

Fraud-detection experts say they are hearing from community colleges in other states that bots are becoming a problem there as well. N2N has been in conversation with colleges in Alabama, Illinois, New Jersey, and Oregon, for example.

Other campuses are keeping fraud detection in house.

They have to be extremely organized, sophisticated, and advanced to be able to do this and not be shut down at this point.

MiraCosta College created “the fraud squad,” which includes representatives from instruction, IT, admissions and records, and financial aid. They created a robust firewall to keep out bots, says Alketa Wojcik, vice president for student services. While the college doesn’t have firm data on how many bots are getting into classrooms, “it hasn’t been in the hundreds,” she says. That’s significant, considering that about one third of their applicants have proved fraudulent.

Denee Pescarmona, vice president for instruction, says it remains critical that administrators stay in close contact with faculty members. If a flood of applicants slip in during the last two minutes before registration closes, she says, the fraud squad’s protocols might not catch them.

There’s also a question about federal oversight, or the lack thereof. Since last October, the U.S Department of Education’s Office of Inspector General has seen its staff shrink by more than 20 percent through buyouts and early retirement, a spokesperson told CalMatters in April. That includes experienced auditors and investigators, she said.

The problem has also gotten caught up in party politics.

Several U.S. Republican representatives in California recently asked the Departments of Education and Justice to investigate the fraud and colleges’ attempts to prevent it as part of the Trump administration’s efforts to “curb wasteful federal spending,” CalMatters reported. And a Democratic state legislator has asked for an audit, noting the effect that the fraud is having on faculty working conditions, student access, and campus resources.

A number of campuses have already noticed what Weber experienced in her courses this spring: fake students emailing profesors directly.

“They are getting smarter,” says Pescarmona, the MiraCosta College administrator. “This isn’t just bots registering. There are folks who are perpetuating the fraud by sending emails to faculty, by requesting add codes, by requesting not to be dropped.”

Suponcic, of BM Technologies, says it stands to reason that scammers will continually adapt.

“By allowing the bad actors to initially get their foot in the door at these schools, it’s allowing them to learn all of the processes, the timelines, the workarounds,” he says. “And that’s how they’ve been able to do this at such a high level. They know everything.”